BiMGCL: rumor detection via bi-directional multi-level graph contrastive learning.

Area of Science:

• Artificial Intelligence
• Social Media Analysis
• Computational Linguistics

Background:

• Large language models accelerate rumor generation, challenging social media content authenticity.
• Existing deep learning methods for rumor detection lack robustness and fail to fully utilize structural information.

Purpose of the Study:

• To propose a novel rumor detection framework, BiMGCL, that addresses the limitations of current methods.
• To improve the accuracy and robustness of rumor identification and detection.

Main Methods:

• Modeling rumor propagation structures as fine-grained bi-directional graphs.
• Employing self-supervised contrastive learning at both node and graph levels.
• Utilizing three interpretable bi-directional graph data augmentation strategies.

Main Results:

• BiMGCL demonstrates superior rumor detection performance compared to state-of-the-art methods.
• The framework effectively captures rumor propagation characteristics through bi-directional graph modeling and contrastive learning.

Conclusions:

• BiMGCL offers a robust and effective solution for rumor detection in social media.
• The proposed framework advances the field by integrating structural information and advanced learning techniques.